Evaluations and benchmarks
Evaluations run on each organization’s own telemetry.
AI Detection, available now to select organizations · AI Investigation, early access
How are Detection and Investigation evaluated?
Each one uses the organization’s labels and a scorecard agreed before any result.
See the methodMethod
The team judges the results against criteria agreed before testing begins.
Detection runs beside existing tools on labeled replays and at least one representative week.
Investigation is tested on malicious cases, benign look-alikes, conflicting evidence and missing data.
Latency includes collection, waiting and processing time. The evaluation also checks how missing feeds are reported.
Cost is measured per accepted investigation, and time saved in minutes per case.
Measures
Detection is scored on precision, false positives, duplicates, time to disposition and hardware use. Investigation is scored on its conclusions, unsupported claims and reviewer effort.
Limits
A result depends on the cases selected, the quality of their labels and the data available. Those conditions belong beside the score, so readers can judge what it supports.
